Fractionation of the visuomotor feedback response to ......2013/05/23 · cally, task-relevant...

Fractionation of the visuomotor feedback response to directions of movementand perturbation

David W. Franklin, Sae Franklin, and Daniel M. WolpertComputational and Biological Learning Lab, Department of Engineering, University of Cambridge, Cambridge,United Kingdom

Submitted 23 May 2013; accepted in final form 4 August 2014

Franklin DW, Franklin S, Wolpert DM. Fractionation of thevisuomotor feedback response to directions of movement and pertur-bation. J Neurophysiol 112: 2218–2233, 2014. First published August6, 2014; doi:10.1152/jn.00377.2013.—Recent studies have high-lighted the modulation and control of feedback gains as support foroptimal feedback control. While many experiments contrast feedbackgains across different environments, only a few have demonstrated theappropriate modulation of feedback gains from one movement to thenext. Here we extend previous work by examining whether differentvisuomotor feedback gains can be learned for different directions ofmovement or perturbation directions in the same posture. To do thiswe measure visuomotor responses (involuntary motor responses toshifts in the visual feedback of the hand) during reaching movements.Previous work has demonstrated that these feedback responses can bemodulated depending on the statistical distributions of the environ-ment. Specifically, feedback gains were upregulated for task-relevantenvironments and downregulated for task-irrelevant environments.Using these two statistical distributions, the first experiment examinedwhether these feedback responses could be independently modulatedfor the same limb posture for two directions of movement (same limbposture but on either an inward or outward movement), while thesecond examined whether the feedback responses could modulate,within a single movement, to perturbations to the left or right of thereach. Both experiments demonstrated that visuomotor feedback re-sponses could be learned independently such that the response wasappropriate for the environment. This work demonstrates that feed-back gains can be simultaneously tuned (upregulated and downregu-lated) depending on the state of the body and the environment. Theresults indicate the degree to which feedback responses can befractionated in order to adapt to the world.

adaptive control of reflex magnitude; motor control; online control;reflex modulation; visually guided reaching

RECENT WORK HAS HIGHLIGHTED the ability of the sensorimotorcontrol system to modulate the feedback responses accordingto the environment. For example, feedback gains have beenshown to modulate according to the environmental dynamics(Cluff and Scott 2013; Franklin et al. 2007, 2012; Kimura andGomi 2009), the limb dynamics (Kurtzer et al. 2008; Pruszyn-ski et al. 2011a), the visual statistics (Franklin and Wolpert2008), the shape of the target (Knill et al. 2011; Nashed et al.2012), and when the two limbs manipulate a single object(Diedrichsen 2007; Dimitriou et al. 2012; Omrani et al. 2013).Such modulation has been used to support the theory that theskillful movement occurs through the setting of the appropriatefeedback gains to the task being performed, a theory termedoptimal feedback control (Liu and Todorov 2007; Scott 2004;

Todorov 2004; Todorov and Jordan 2002). Although moststudies have examined differences in the feedback responses ina blocked design (e.g., Ahmadi-Pajouh et al. 2012; Nashed etal. 2012), several papers have recently demonstrated modula-tion of these responses within a single trial (Dimitriou et al.2013; Kimura and Gomi 2009; Knill et al. 2011). Dimitriouand colleagues (2013) demonstrated an online update in thefeedback gain due to changes in the location of the reach target.By using targets with different shapes, Knill et al. (2011)showed that the visuomotor feedback gains can be modulateddifferently for perturbations in the direction of the reachcompared with perturbations laterally. Kimura and Gomi(2009) measured the stretch response to mechanical perturba-tions, showing that magnitude was modulated early in a move-ment depending on the force field that would be applied laterin the movement. This work showed that subjects were able toselect the appropriate stretch feedback gain, different for per-turbations to one side of the movement or the other, based onthe task being performed and to adapt this from one trial to thenext. Here we examined whether such differential modulationof feedback gains for perturbation direction is also possible forvisuomotor perturbations. Although the work of Knill andcolleagues (2011) shows modulation of feedback gains be-tween direction and extent for different target shapes, as thetargets were symmetrical about the two principal axes it wasnot possible to examine whether gains could be modulatedwithin a single axis for two directions of perturbation. Here weinvestigate whether visuomotor feedback gains can also bemodulated independently to the left or right of a reachingmovement for the same limb posture (i.e., for identical limbposture and distance to the target). Such a differential responsewould suggest either that an optimal controller sets indepen-dent feedback gains to errors on one side of the movement orthe other or that the controller switches between multipleoptimal controllers.

Rapid motor responses to visual signals occur in response tothe presentation of a visual stimulus (Corneil et al. 2004, 2007)and shifts in the target location (Prablanc and Martin 1992), theentire background (Saijo et al. 2005), and the representation ofthe hand position (Brenner and Smeets 2003; Sarlegna et al.2003; Saunders and Knill 2003). The visually induced correc-tive responses occur relatively quickly (150 ms) after therepresentation of the hand or target moves and do not requiresubjects to consciously perceive these movements (Goodale etal. 1986; Prablanc and Martin 1992). Early components of allof these visually induced motor responses have been shown tobe involuntary in nature (Day and Lyon 2000; Franklin andWolpert 2008; Gomi et al. 2006; Saijo et al. 2005).

Address for reprint requests and other correspondence: D. W. Franklin,Dept. of Engineering, Univ. of Cambridge, Trumpington St., Cambridge, CB21PZ, UK (e-mail: [email protected]).

J Neurophysiol 112: 2218–2233, 2014.First published August 6, 2014; doi:10.1152/jn.00377.2013.

2218 Licensed under Creative Commons Attribution CC-BY 3.0: © the American Physiological Society. ISSN 0022-3077. www.jn.org

on Novem

ber 2, 2014D

ownloaded from

mailto:[email protected]

http://creativecommons.org/licenses/by/3.0/deed.en_US

Previously we showed that this visuomotor reflex can betuned to the statistical properties of the environment (Franklinand Wolpert 2008). In particular, we found an increase in themagnitude of the response when the environment containedtask-relevant sensory discrepancies and a reduction in gainwhen the environment contained task-irrelevant sensory dis-crepancies. The sensory discrepancy environments were pro-duced by smoothly moving the visual feedback away from theactual hand position in the middle of the movement. In thetask-relevant condition, these discrepancies were maintainedand therefore required correction, while in the task-irrelevantcondition these discrepancies were smoothly removed suchthat the visual feedback was matched with the actual handposition at the end of the movement. In the present study weexploit these findings to examine whether the sensorimotorcontrol system can control the magnitude or gain of the reflexresponse simultaneously to both task-irrelevant and task-rele-vant sensory discrepancies. First we determine whether thevisuomotor reflex can be tuned differently in the same spatiallocation but with movements of different directions. Specifi-cally, task-relevant discrepancies are applied for one directionof movement and task-irrelevant discrepancies for the other.Second, we examine whether the feedback responses can betuned independently to perturbations on the right and on theleft of the reaching movement by varying the task relevancy ofsensory discrepancies on either side of the reaching movement(with task-irrelevant discrepancies on one side and task-rele-vant discrepancies on the other). We demonstrate that thesensorimotor control system can tune the reflex gain simulta-neously in a particular limb posture to different states of thelimb prior to the perturbation and to different perturbationdirections.

MATERIALS AND METHODS

Twenty-four subjects participated in 2 groups of 12 subjects for thefirst experiment: group 1 (4 women and 8 men; mean ! SD age:23.9 ! 5.8 yr) and group 2 (7 women and 5 men; age: 23.8 ! 4.8 yr).Sixteen subjects (10 men and 6 women; mean age: 25.1 ! 6.5 yr), sixof whom also participated in the first experiment, were recruited toparticipate in the second experiment. All subjects were right-handedaccording to the Edinburgh handedness inventory (Oldfield 1971),with no reported neurological disorders and normal or corrected tonormal vision. Subjects gave informed consent, and the experimentswere approved by the institutional ethics committee.

Experimental setup. Movements investigated in this study wereright-handed outward (away from the body) and inward (toward thebody) movements in the horizontal plane at "10 cm below thesubjects’ shoulder level. The forearm was supported against gravitywith an airsled. The handle of the robotic manipulandum (vBOT) usedto generate the environmental dynamics (null field or mechanicalchannels) was grasped by the subject (Fig. 1A). Position and forcedata were sampled at 1 kHz. End-point forces at the handle weremeasured with an ATI Nano 25 six-axis force-torque transducer (ATIIndustrial Automation). Visual feedback was provided with a com-puter monitor mounted above the vBOT and projected veridically tothe subject via a mirror. This virtual reality system covers themanipulandum, arm, and hand of the subject, preventing any visualinformation of their location. Full details of the vBOT have beenpublished previously (Howard et al. 2009). The exact time that thestimuli were presented visually to the subjects was determined byusing the video card refresh rate and confirmed with an optical sensor.

Movements were made from a 1.0-cm-diameter start circle to a2.0-cm-diameter target circle, both of which were centered in front of

the subject. The subjects’ arm was hidden from view by the virtualreality visual system on which the start and target circles, as well asa 0.6-cm-diameter cursor used to track instantaneous hand position,were projected. The distance between the centers of the start andtarget circles was 25.0 cm. Successful movements were defined asthose that entered the target without overshooting and with movementdurations in the range 700 ! 75 ms and were accompanied withfeedback (“good” or “great”). When subjects performed successfulmovements, a counter increased, and subjects were instructed toincrease the number of the counter as much as possible throughout theexperiment. When subjects performed unsuccessful movements, theywere provided with feedback as to why the movement was notconsidered successful (“too fast,” “too slow,” or “overshot target”).Trials were self paced; subjects initiated a trial by moving the handcursor into the start circle and holding it within the target for 450 ms.A beep then indicated that the subject could begin the movement tothe target. The duration of the movement was determined from thetime that the subject’s hand exited the start target until the time thatthe subject’s hand entered the final target.

Electromyography. In the second experiment, surface electromyog-raphy (EMG) was recorded from two monoarticular shoulder muscles:pectoralis major and posterior deltoid; two biarticular muscles: bicepsbrachii and long head of the triceps; and two monoarticular elbowmuscles: brachioradialis and lateral head of the triceps. EMG wasrecorded with the Delsys Bagnoli (DE-2.1 Single Differential Elec-trodes) EMG system (Boston, MA). The electrode locations werechosen to maximize the signal from a particular muscle while avoid-ing cross talk from other muscles. The skin was cleansed with alcoholand prepared by rubbing an abrasive gel into the skin. This wasremoved with a cotton pad, and the gelled electrodes were secured tothe skin with double-sided tape. The EMG signals were band-passfiltered online through the EMG system (20–450 Hz) and sampled at2 kHz. The EMG signals are aligned with the force and positionsignals with a signal from the serial port, which is recorded on achannel with the EMG. This signal is changed at the onset and offsetof any perturbation in order to ensure that the correct timing isachieved over the period of interest.

Probe trials to measure reflex gain. Visually induced motor re-sponses were examined with perturbations similar to those previouslydescribed (Dimitriou et al. 2013; Franklin et al. 2012; Franklin andWolpert 2008). In the middle of the movements to the target, thecursor representing the hand position was laterally jumped away fromthe current hand position, held at a fixed distance from the actual handtrajectory for 250 ms, and then returned to the actual hand position forthe rest of the movement. The entire visual perturbation lasted for 250ms. During such probe trials, the hand was physically constrained tothe straight path between the start and final targets [mechanicalchannel trial (Milner and Franklin 2005; Scheidt et al. 2000) generatedby the vBOT]. The mechanical channel was implemented as a stiff-ness of 5,000 N/m and damping of 2 N·m#1·s#1 for any movementlateral to the straight line joining the starting location and the middleof the target. This constrains the physical hand location using thechannel such that no change in the arm configuration occurs. Theforce produced in response to the visual perturbation can be measuredagainst the channel wall with the force sensor. As the visual pertur-bation returns to the actual hand trajectory, subjects do not need torespond to the visual perturbation in order to produce a successfulmovement to the target. A discussion of the possible concerns with theuse of mechanical channels to measure visuomotor feedback gains canbe found in the methods section of Dimitriou et al. (2013). Thesevisual perturbations were applied perpendicular to the direction of themovement (either to the left or the right). For comparison, a zero-perturbation trial was also included, in which the hand was held to astraight-line trajectory to the target but the visual cursor remained atthe hand position throughout the trial. The perturbation trials wererandomly applied during movements in a blocked fashion such that

2219FRACTIONATION OF VISUOMOTOR FEEDBACK RESPONSE

J Neurophysiol • doi:10.1152/jn.00377.2013 • www.jn.org

on Novem

ber 2, 2014D

ownloaded from

one of each perturbation type (each perturbation direction or onsettime) was applied within each block of trials.

Experiment 1—effect of movement direction. Our previous exper-iment (Franklin and Wolpert 2008) demonstrated that subjects couldadapt the magnitude of the reflex response (either increasing ordecreasing the reflex response) to two different distributions of sen-sory discrepancies (task relevant or task irrelevant, respectively) whenthey were presented on different days. Here we examine whethersubjects can simultaneously control their reflex response to the twodifferent distributions when they are presented in the same locationbut for different directions of movement (Fig. 1B). Subjects madeboth outward and inward movements directly in front of their body.There were two stages to the experiment. The first half of theexperiment (comprised of 1,122 movements) was performed in anormal environment, one in there was no discrepancy between thevisual path of the hand and the actual physical path of the hand. Forthe first group of subjects (n $ 12), the second half of the experiment

had task-relevant discrepancies on the outward movements and task-irrelevant discrepancies on the inward movements. A second group ofsubjects (n $ 12) had the reverse: task-irrelevant discrepancies on theoutward movements and task-relevant discrepancies on the inwardmovements. In the task-relevant condition, at a point in the trajectorythat was 30% of the distance to the target (7.5 cm from the start), thevisual cursor representing the hand position underwent a smooth(minimum jerk) movement laterally to the movement direction to adistance from the set [#5.0, #3.75, #2.5, #1.25, 0, 1.25, 2.5, 3.75,5.0] cm in the subsequent 8.5 cm of the movement distance, remainingat this position laterally for the rest of the movement (Fig. 1C, left).Subjects were required to determine the appropriate response in orderto bring the hand cursor back in to the target and be credited with asuccessful trial. In this condition, while the lateral change in the visuallocation of the hand position produces a visuomotor discrepancy, thevisual signal provides reliable information about the amount ofcompensation that the subjects will need to produce in order tosuccessfully complete the movement. In the task-irrelevant condition,30% of the distance to the final target (7.5 cm from the start), thevisual cursor representing the hand position underwent a smooth(minimum jerk) movement laterally to a distance from the set [#5.0,#3.75, #2.5, #1.25, 0, 1.25, 2.5, 3.75, 5.0] cm in the subsequent 8.5cm of movement and then returned to the actual hand position in thefinal 8.5 cm of outward movement in the same manner (Fig. 1C,right). Thus at the final target the hand position and cursor positionwere matched. While this condition initially produces a visuomotordiscrepancy identical to the task-relevant condition, the size of thisdiscrepancy between visual and haptic signals in the middle of the

-1.5 0 1.5

0

25

-5 0 5-2 0 2 -5 0 5-2 0 2

B C

Posi

tion

Y-ax

is [c

m]

Position X-axis [cm]-1.5 0 1.5

D probe trialsnormal sensory discrepancyenvironments

E

A

GF probe trialsnormal sensory discrepancyenvironments

-5 0 5 -2.5 0 2.5-5 0 5-5 0 5Position X-axis [cm]

0

25

Posi

tion

Y-ax

is [c

m]

vBOT

airsled

target

starty

x[0,0]

vBOT

monitor

airsled table

mirror

Fig. 1. The experimental setup. A: the subject grasps the robotic manipulandum(vBOT) while seated. Visual feedback is presented veridically with a top-mounted computer screen viewed through a mirror. The subject’s forearm isfixed to and supported by an airsled. B: experiment 1: effect of movementdirection. Subjects made movements alternately in 2 directions (outward andinward). Initially, movements were performed in the normal condition, wherethe cursor reproduced the hand trajectory exactly. Note that for clarity ofdisplay the outward and inward movements have been horizontally offset inthe plots. C: in the second phase of the experiment, the outward movement wasperformed in the task-relevant environment (red-yellow) whereas the returnmovements were performed in the task-irrelevant environment (cyan-blue). Asecond group of subjects performed experiments in which these directionswere reversed. In the task-relevant condition, the visual cursor smoothlymoved away from the hand trajectory to 1 of 9 amplitudes (including 0) andremained at this point for the rest of the movement. In the task-irrelevantcondition, the visual cursor smoothly moved away from the hand trajectoryexactly as in the task-relevant condition but then returned smoothly such thatit agreed with the physical hand position at the end of the movement. D: thevisual perturbations (probe trials) used in experiment 1 to examine themagnitude of the visually induced motor response. On random trials, the handwas mechanically constrained to a straight-line trajectory to the target and thevisual cursor representing the hand was jumped laterally away from the actualhand position and returned 250 ms later. Two different onsets of the pertur-bations (light green and orange arrows) were chosen for each direction: 1 pair(orange) were matched for the spatial location of the visual perturbation of the2 directions of movement (matched stimuli perturbations: occurring at 30% ofthe distance to the target), and the other pair (light green) were matched so thatthe visuomotor response occurred at a similar location for the 2 directions ofmovement (matched response perturbations: occurring at 10% of the distanceto the target). E: experiment 2: effect of perturbation direction. Subjectsperformed outward reaching movements. Initially, all subjects performedmovements in the normal environment. F: in the second half of the experiment1 group of subject made movements under 1 of the sensory discrepancyenvironments (left), whereas the other group was presented with the oppositeenvironment (right). In these environments, the visual cursor representing thesubject’s hand was manipulated in a task-relevant manner on one side of themovement and in a task-irrelevant manner on the other side of the movement.G: the visual perturbations (probe trials) used in experiment 2 to examine themagnitude of the visually induced motor response. Two sizes of perturbationwere used on either side of the zero perturbation to examine changes in thegain of the response.

2220 FRACTIONATION OF VISUOMOTOR FEEDBACK RESPONSE


on Novem

ber 2, 2014D

ownloaded from

movement does not provide reliable information about the location ofthe hand at the end of the trial. After each movement, the roboticmanipulandum moved the subjects’ limb to the start position, whilevisual feedback of the location of the hand was not presented.

In all environments five different types of visual perturbation trialsor probe trials (Fig. 1D) were presented within a single block of 14trials (5 probe trials and 9 nonprobe trials from the distribution) ineach direction in order to assess the feedback gains. This means thata single block of movements (28 movements) was made up of 14outward movements and 14 inward movements. If the nine nonprobemovements were from a distribution of sensory discrepancies, thenone of each discrepancy was included in the block of movements.Four of the five probe trials used in this experiment were two 1.5-cmvisual displacements either to the left or the right of the movementoccurring at either 10% or 30% of the movement distance and lastingfor 250 ms. The fifth probe trial had a zero visual displacement forcomparison against the shifted visual perturbations. The perturbationsoccurring at 10% of the movement distance were implemented toexamine the response magnitude when the positions at which the forceresponse occurs during the movement were matched for both theoutward and inward reaching movements. On the other hand, theperturbations occurring at 30% of the movement distance were im-plemented to examine the responses when the stimuli (the visualperturbations) were matched for the limb posture in the inward andoutward movements. A single nonprobe movement was always per-formed first in any new environment, such that a probe trial was neverthe first movement. While lateral movement in the random probe trialswas constrained by the mechanical channel, the subjects were free tomove in any manner during all of the other trials. To examine theeffects of each of the two changed conditions (task relevant or taskirrelevant), the responses were contrasted with the normal condition inthe same movement direction in order to get an appropriate baselinefor both conditions. Subjects performed 40 blocks of movements ineach condition. This resulted in 1,122 movements in the normalcondition (561 outward and 561 inward) followed by 1,122 move-ments in the sensory discrepancy environments (561 outward taskrelevant and 561 inward task irrelevant or vice versa for the secondgroup of subjects). Subjects were required to take short breaks every300 movements throughout the experiment. They were also allowed torest at any point they wished by releasing a safety switch on thehandle.

Experiment 2—effect of perturbation direction. The second exper-iment was designed to examine whether the reflex response could bemodulated differently for different directions of visual perturbation.To determine whether the reflex gains could be modulated simulta-neously, subjects were exposed to a distribution of sensory discrep-ancies that were all task relevant on one side and task irrelevant on theother side of the reaching movement (Fig. 1F). Subjects performed theexperiment on two separate days. All subjects started with the normalenvironment in the first phase of the experiment (Fig. 1E); the secondphase was performed differently for two randomly chosen groups. Onday 1, half of the subjects experienced the task-relevant sensorydiscrepancies on the right side of the reaching movement (positiveperturbations) and the task-irrelevant sensory discrepancies on the leftside of the reaching movement (negative perturbations) while theother half experienced the opposite mapping of direction to taskrelevance (Fig. 1F). On day 2 the mapping of direction to taskrelevance was reversed for all subjects.

Subjects made outward reaching movements to a target centereddirectly in front of their body. Each day of the experiment had twocomponents. Initially, subjects made reaching movements in a normalenvironment—one in which the physical location and the visuallocation of the hand matched throughout the movement. After 50blocks of this condition were completed, the sensory discrepancyenvironment was presented. In this environment, at a point in thetrajectory that was 40% of the distance to the target (10 cm from thestart), the visual cursor representing the hand position underwent a

smooth (minimum jerk) movement laterally to the movement direc-tion to a distance from the set [#5.0, #3.75, #2.5, #1.25, 0, 1.25,2.5, 3.75, 5.0] cm in the next 7.5 cm of the outward movementdistance. On perturbations to one side, the displacement was main-tained throughout the rest of the movement, requiring subjects toactively respond to this displacement and move the visual location ofthe hand back toward the target in order to achieve the task. However,for perturbations to the other side, the visual signal of the hand wasreturned to the actual hand position over the final 7.5 cm of outwardmovement in the same manner such that at the final target the handposition and cursor position were matched. Note that in both thesensory discrepancy environment and the normal environment, thereare trials in which the visual cursor perfectly matches the handlocation during the reach (zero sensory discrepancy trials). Thesetrials can be used to examine whether there are any differences in thedirection and curvature of the movement between the two conditions.

During both phases of the experiment, five different visual pertur-bation trials or probe trials (Fig. 1G) were presented within a singleblock of 14 trials (5 probe trials and 9 unperturbed trials from thedistribution) in each direction in order to assess the reflex response.The five probe trials consisted of two perturbations to the left [#1.25cm, #2.5 cm], two perturbations to the right [1.25 cm, 2.5 cm], anda zero perturbation for comparison. The perturbations were initiated at45% of the distance to the target (11.25 cm into the movement) andlasted for 250 ms. The other nine unperturbed trials contained in eachblock were either composed of nine normal trials (normal environ-ment) or composed of one of each of the sensory discrepancy trialssuch that every block contained equal numbers of both the task-relevant and task-irrelevant trials (sensory discrepancy environment).A single movement was always performed first in any new environ-ment, such that a probe trial was never the first movement. Whilelateral movement in the random probe trials was constrained by themechanical channel, the subjects were free to move in any mannerduring all of the other trials. Subjects performed 50 blocks of move-ments in each phase. This resulted in 701 movements in the normalcondition followed by 701 movements in the sensory discrepancyenvironment. Subjects were required to take short breaks every 300movements throughout the experiment. They were also allowed to restat any point they wished by releasing a safety switch on the handle.

Analysis. Analysis of the experimental data was performed withMATLAB R2012a. EMG data were high-pass filtered with a fifth-order, zero phase-lag Butterworth filter with a 30-Hz cutoff and thenrectified. Hand acceleration was obtained by double differentiating theposition data and low-pass filtering at 15 Hz with a fifth-order, zerophase-lag Butterworth filter after each differentiation. As performedpreviously (Dimitriou et al. 2013; Franklin and Wolpert 2008), theresponses to visual perturbations in the first five blocks of a conditionwere not used for analysis in order to avoid the possible influence ofinitial high-gain trials at the beginning of the experiments. Thereforeonly the results from the last 45 blocks (blocks 6–50 ) are used in anyof the analyses and results. Individual probe trials were aligned onvisual perturbation onset and averaged across repetitions. The re-sponse to the right visual perturbation on probe trials was subtractedfrom the response to the left perturbation on probe trials in order toprovide a single estimate of the motor response to the visual pertur-bation for the first experiment. In the second experiment, the re-sponses to the right and left perturbations were examined separatelyby subtracting them from the zero perturbation. In this manner theresponse to each side could be examined. To examine the rapidresponse (i.e., “reflex”), we calculated the average postperturbationEMG (120–180 ms) and force (180–230 ms) (Franklin and Wolpert2008). ANOVAs were examined in SPSS (v. 21) with the generallinear model. If a significant main effect was found, Tukey’s honestlysignificant difference (HSD) post hoc test was used to examinedifferences. Statistical significance was considered at the P % 0.05level for all statistical tests. Specifically, each data point used forstatistical analysis represented the average across the reflex time



on Novem

ber 2, 2014D

ownloaded from

window of a single perturbation trial for EMG and the differencebetween pairs of trials for force. All statistical tests on the EMG dataare performed on unscaled and unsmoothed subject data. For plottingpurposes only (Fig. 7), the EMG was scaled, smoothed, and averagedacross subjects (but not across muscles). To do this for a particularmuscle we calculate a single scalar for each subject that is used toscale the muscle’s EMG traces for all trials for that subject. The scalarwas chosen so that the mean (across trials) of the EMG data averagedover the period #50 to 50 ms relative to the onset of the perturbationwas equal across subjects (and set to the mean over all the subjects).This puts each subject on an equal scale to influence any responseseen in the data. Prior to averaging, the individual muscle traces wereinitially smoothed with a 10-point (5 ms) moving average. Onceaveraged across all subjects, the mean and SE were determined foreach condition and the mean muscle activity was then smoothed witha 10-point (5 ms) moving average prior to plotting.

For both experiments 1 and 2, analysis was performed in order todetermine the earliest time at which visuomotor responses weremodulated independently for the two environments. Specifically, toexamine whether there was independent modulation of the feedbackresponses for different states and to determine at what time suchindependent modulation occurs, the difference in the force response inthe sensory discrepancy environment was calculated every 1 msrelative to the temporal pattern in the normal environment (mean foreach subject). t-Tests were then used to test whether the difference inthese responses was significant between the task-relevant and task-irrelevant environments at each time point. Significant differenceswere considered at P % 0.05, Bonferroni corrected for multiplecomparisons.

RESULTS

Experiment 1—effect of movement direction. Subjects per-formed reaching movements while holding the handle of arobotic manipulandum (Fig. 1A). On random probe trials dur-ing normal reaching movements subjects were presented witha visual perturbation of the hand cursor while the hand itselfwas mechanically constrained to move within a channel to thetarget. These perturbations occurred either to the left or right ofthe hand trajectory at either 10% of the movement distance or30% of the movement distance (Fig. 1D) in order to produceperturbations occurring at the same physical location on theoutward and inward movements. The early perturbations pro-duced a feedback response as measured by the force exertedagainst the channel wall occurring at the same physical loca-tion for both the outward and inward movements (Fig. 2B). Inthis condition, the physical location of the arm was similar inboth directions of movements at the time of the feedbackresponse. On the other hand, in case the onset of the perturba-tion was the relevant state to determine the feedback gain, thelater perturbation led to the perturbation occurring at the samelocation for the outward and inward movements (Fig. 2A). Byusing both perturbations, the modulation of the involuntaryvisuomotor responses could be examined for matching physi-cal locations of the arm both for the visual stimuli and for thereflex response to determine whether the feedback responsescould be independently tuned for different reaching velocities(negative or positive) for the same physical location.

In the initial half of the experiment (1,122 movements), bothdirections of movement were performed in the normal envi-ronment where the actual position of the hand was accuratelyrepresented by the visual feedback in all nonprobe trials. In thesecond half of the experiment, the outward reaching move-ments were performed in a task-relevant sensory discrepancy

environment while the inward reaching movements were per-formed in a task-irrelevant sensory discrepancy environment(group 1; Fig. 1C) or vice versa (group 2). The hand kinemat-ics and visual feedback for these movements are shown in Fig.3. In the task-relevant environment subjects made correctiveresponses late in the movement to reach the final target (Fig.3B), whereas in the task-irrelevant environment subjects didnot respond to the visual motion, instead performing straightmovements to the target (Fig. 3D).

The force responses to the visual perturbations (probe trials)were compared between the two stages of the experiments forthe same reaching direction. For the first group of subjects, theforce responses were relatively low in the normal environmentbut increased rapidly to a high level in the task-relevantsensory discrepancy environment for the outward reaching

0

0

0

0

0

0

0

0

Cur

sor P

ositi

on [c

m]

Forc

e X-

axis

[N]

Cur

sor P

ositi

on [c

m]

Forc

e X-

axis

[N]

Matched Stimuli Position Matched Response PositionA B

direction ofmovement




Position Y-axis [cm] Position Y-axis [cm]

Fig. 2. Comparison of the timing of perturbations in experiment 1. For eachmovement direction, the cursor position (x-axis) and the corresponding changein force against the channel wall (x-axis) are plotted as a function of theposition in the y-axis. Mean and SE (shaded region) are plotted across all 24subjects from experiment 1. A: forces during the 30% onset cursor perturba-tions (matched stimuli position) for the outward reaching (top) and inwardreaching (bottom) movements. Shaded gray region demonstrates the matchedpositions (limb postures) over which the cursor perturbations in the outwardand inward movements occur. B: forces during the 10% onset cursor pertur-bations (matched response position) for the outward reaching (top) and inwardreaching (bottom) movements. Shaded gray region demonstrates the matchedpositions (limb postures) over which the force responses to the visuomotorperturbation in the outward and inward movements occur.



on Novem

ber 2, 2014D

ownloaded from

movements (Fig. 4A). The visuomotor responses were com-pared in the involuntary response interval (180–230 ms) toexamine whether differences could be distinguished prior tothe possible involvement of voluntary feedback changes (Fig.4B). For both the 10% (matched response) and 30% (matchedstimuli) onset perturbations, the forces were higher in thesensory discrepancy environment. These differences were ex-amined with an ANOVA with main effect of environment typeand subject as a random effect. The responses were found tobe significantly larger in the task-relevant condition matchedfor both 10% onset perturbation (F1,11 $ 13.805; P $ 0.003)and 30% onset perturbation (F1,11 $ 103.311; P % 0.001)locations.

Similarly, the responses were examined for the inwardreaching movements (Fig. 4, C and D). While the initial forceresponses were similar in both the normal and sensory discrep-ancy environments, the response decreased in the task-irrele-vant environment "300 ms after the onset of the visualperturbation (Fig. 4C). Again the responses were comparedduring the involuntary response time (180–230 ms) (Fig. 4D).By ANOVA, the responses were found to be not significantlydifferent between the two conditions for both 10% onsetperturbation (F1,11 $ 1.151; P $ 0.306) and 30% onsetperturbation (F1,11 $ 0.47; P $ 0.507) locations.

The results from the above analysis suggest that the feed-back forces were modulated independently for the outward andinward directions within the involuntary time window. Toexamine the time point at which such independent modulationoccurs in the feedback responses, the difference in the forceresponses between the final sensory discrepancy environmentand the initial normal environment was calculated every 1 msthroughout the movement for each subject (Fig. 4, E and F).t-Tests (Bonferroni corrected for multiple comparisons) werethen used to test whether the difference in these responsesbetween the outward movements (task relevant) and inwardmovements (task irrelevant) was significant at each time point.Significant differences in the responses were found startingfrom 189 ms (t834 $ 3.803; P $ 1.53e#04) for the 10% onset

perturbations and from 179 ms (t838 $ 3.971; P $ 7.76e#05)for the 30% onset perturbations, well before the earliest vol-untary change in feedback force (230 ms) for such visualperturbations (Franklin and Wolpert 2008). It is important tonote that the onset times for the excitation of the response inthe task-relevant environment (relative to the normal environ-ment) is much earlier (10% onset: 212 ms; 30% onset: 178 ms)than the inhibition of the response found in the task-irrelevantenvironment (10% onset: 298 ms; 30% onset: 274 ms).

A second group of subjects performed the reverse experi-ment where the task-irrelevant sensory discrepancy occurredon the outward movement and the task-relevant discrepancy onthe inward movement in the second half of the experiment. Theresults showed the same effects across all conditions as forgroup 1. The responses were found not to be significantlydifferent in the task-irrelevant condition matched for either10% onset (F1,11 $ 3.059; P $ 0.108) or 30% onset (F1,11 $0.081; P $ 0.782) perturbations (Fig. 4, G and H). However,the responses were found to be significantly larger in thetask-relevant condition for both the 10% onset (F1,11 $ 15.835;P $ 0.002) and 30% onset (F1,11 $ 34.56; P % 0.001)perturbations (Fig. 4, I and J). Again, t-tests (Bonferronicorrected for multiple comparisons) were used to determinewhether there were significant differences in these responsesbetween the outward movements (task irrelevant) and inwardmovements (task relevant). Significant differences in the re-sponses were found starting from 201 ms (t838 $ 3.84; P $1.32e#4) for the 10% onset perturbations (Fig. 4K) and from185 ms (t838 $ 3.94; P $ 8.89e#5) for the 30% onset pertur-bations (Fig. 4L), again well before the earliest voluntaryresponse (230 ms) for such visual perturbations (Franklin andWolpert 2008). Note again that these early differences arebrought about by the excitation in the task-relevant (comparedto the normal) condition rather than the inhibition in thetask-irrelevant (compared to the normal) condition. The onsettimes in the task-relevant conditions were 172 ms (10% onset)and 174 ms (30% onset), whereas in the task-irrelevant theywere 266 ms and 247 ms, respectively.

Experiment 2—effect of perturbation direction. In the sec-ond experiment, subjects made outward reaching movementsto a target. During the first half of the experiment subjectsperformed these movements in the normal environment wherethe visual feedback matched the physical position of the hand(Fig. 1E), while during the second half of the experiment asensory discrepancy environment was introduced (Fig. 1F).This sensory discrepancy environment was task relevant onone side of the reaching movement and task irrelevant on theother side (subjects experienced both variants of the sensorydiscrepancy environment on separate days, counterbalancedacross subjects). The hand kinematics and visual feedback forthese movements are shown in Fig. 6. Similar to experiment 1,when subjects had movements in which the visual path wastask relevant they made corrective responses late in the move-ment to reach the final target, whereas when the visual pathwas task irrelevant subjects did not respond to the visualmotion, instead performing straight movements to the target(Fig. 5, B and D).

The movements on these sensory discrepancy trials wereexamined to see any evidence for a change in the feedbackresponses (Fig. 6). The lateral hand acceleration changed "150ms after the onset of the sensory discrepancies, with larger and

-5 0 5 -5 0 5 -5 0 5

0

25

-5 0 5

DCA B

Posi

tion

Y-ax

is [c

m]

Position X-axis [cm] Position X-axis [cm]

Fig. 3. Kinematics in experiment 1 for group 1. A: mean cursor trajectories forthe 9 different trajectories for a single subject in the task-relevant conditionduring outward reaching movements. B: mean hand trajectories for the samemovements are shown for the same subject, with the colors indicating thecorresponding cursor motions. C: mean cursor trajectories for the 9 differenttrajectories for the same subject in the task-irrelevant condition during inwardreaching movements. D: mean hand trajectories for the same movements.



on Novem

ber 2, 2014D

ownloaded from

more scaled responses to the task-relevant discrepancies thanto the task-irrelevant discrepancies (Fig. 6, B and F). Suchresponses suggest that the subjects were modulating theirfeedback gains separately to either side of the reaching move-ment. This was examined in more detail by determining thedifference in the size of the response for sensory discrepanciesof matched magnitude but opposite sign (e.g., &5 vs. #5)relative to the undisturbed trajectory (Fig. 6, D and H). If themagnitudes of the acceleration were identical then the differ-ence would remain at zero; however, for both cases the re-sponse in the direction of the task-relevant environment was

significantly higher than for the task-irrelevant environment.The onset of the differences was examined with t-tests of theacceleration responses at each point in time, with significancedetermined at P % 0.05 (Bonferroni corrected for multiplecomparisons). Seven of the eight comparisons showed onsetsof the modulation of the responses occurring prior to theearliest voluntary response in force (230 ms; Franklin andWolpert 2008). However, another modulation between theenvironments could also be seen. Specifically, the trajectory inthe zero sensory discrepancy trials shifted between the initialno sensory discrepancy environment and the final sensory

0 100 200 300 400

Forc

e [N

] (le

ft - r

ight

)

Forc

e (1

80-2

50m

s) [N

] (le

ft - r

ight

)

Time after perturbationonset [ms]

G H

10% Onset 30% OnsetN SD N SD

0 100 200 300 400

Forc

e [N

] (le

ft - r

ight

)

Forc

e (1

80-2

50m

s) [N

] (le

ft - r

ight

)


I J


0

1

0

1

0

1

2

2

0

0 100 200 300 400

Forc

e [N

] (le

ft - r

ight

)

Forc

e (1

80-2

50m

s) [N

] (le

ft - r

ight

)


A B

10 % Onset 30% OnsetN SD N SD

0 100 200 300 400

Forc

e [N

] (le

ft - r

ight

)

Forc

e (1

80-2

50m

s) [N

] (le

ft - r

ight

)


C D


0

1

2

0

1

0

1

2

0

1 1

***

**

***

**

Forc

e [N

]p-

valu

es

Forc

e [N

]p-

valu

es

E F10% Onset 30% Onset

0

0.6

-0.6

0

0.6

-0.6

100

10-20

100

10-20

100 200 300 400Time after perturbation

onset [ms]

100 200 300 400

100 400 100 400


p=1.7x10-4 p=1.7x10-4

Forc

e [N

]p-

valu

es

Forc

e [N

]p-

valu

es

K L10% Onset 30% Onset

0

1

-1

0

1

-1

100

10-20 10-20


onset [ms]

100 200 300 400

100 400 100 400


100

189

298

212

274

178179 201

266172

185

247174



on Novem

ber 2, 2014D

ownloaded from

discrepancy environment. This was examined in detail over theentire movement, using both the trajectory on the identical zerosensory discrepancy trials and the lateral force on the zeroperturbation channel trials (Fig. 6, G and H). Subjects shiftedtheir mean trajectory by 6.2 mm (at the end of the movement)toward the task-irrelevant side of the environment. This adap-tation of the mean trajectory has two consequences, the first toassist the correction of the task-relevant shift in the cursorposition and the second to compensate (in advance) for theincompletely suppressed responses to the task-irrelevant shiftsin the cursor position. Thus a shift in the mean trajectory assistswith the compensation to this environment. This trajectoryshift was also apparent on the force exerted on the channeltrials with mean differences of 0.26 N between the twoconditions. Together, these results demonstrate that subjectsadapted to the environment, at least partially, through trajec-tory modulation and also suggest that they modulated theirfeedback responses. However, these sensory discrepancy trialsprovide no estimate of the baseline feedback gain in the normaltrials and also introduce difficulty in determining the detectableonset time of the cursor shift because of their smooth onset.Therefore we use an independent measure of feedback re-

sponses on probe trials to assess any changes in the feedbackgains.

On random trials, in order to estimate the feedback gains, avisual perturbation of the cursor occurred for 250 ms in whichthe location of the cursor was shifted to either the left or rightby 1.25 or 2.5 cm (probe trials) (Fig. 1G). These perturbationswere used to examine whether subjects could independentlyregulate the gain of the feedback responses to the left and rightof the reaching movement on the same movement. The forceresponses to the visual perturbations (probe trials) were com-pared between the two stages of the experiments for eachvariant of the sensory discrepancy environment separately. Theforce responses to the visual perturbation during the normalenvironment were similar in size to the right and left of themovements (Fig. 7, A and E), with the response to the 2.5-cmperturbation slightly larger than that to the 1.25-cm perturba-tion. Once the sensory discrepancy environment was intro-duced, perturbations into the task-relevant side of the environ-ment produced larger force responses for both the 1.25-cmperturbation and 2.5-cm perturbation (Fig. 7, A and E). On theother hand, perturbations into the task-irrelevant side of theenvironment produced little change in the force response foreither the 1.25-cm or the 2.5-cm perturbation.

The visuomotor responses were compared in the involuntaryresponse interval (180–230 ms) to examine whether differ-ences could be distinguished prior to the possible involvementof voluntary feedback changes (Fig. 7, B and F). The magni-tudes of the force responses were compared separately forperturbations to the right and left. On the experimental day inwhich the task-relevant sensory discrepancy occurred on theleft hand side of the straight reaching movement (Fig. 7, A andB), an ANOVA with main effects of perturbation size (2 levels:1.25 cm and 2.5 cm) and environment (2 levels: normal andsensory discrepancy) and random effect of subject was per-formed for the perturbation to the left (positive force responsesin Fig. 7A). Significant differences were found for the size ofthe perturbation (F1,15 $ 93.714; P % 0.001) and the environ-mental condition (F1,15 $ 15.161; P $ 0.001) and an interac-tion between these two (F1,15 $ 14.374; P $ 0.002), demon-strating that the task-relevant environment increases the gain ofthe visuomotor responses. Finally, for direct comparisonsacross the conditions an ANOVA was run with a main effect ofcondition (4 levels) and random effects of subjects and plannedcomparisons between conditions either with the same pertur-

Fig. 4. Visuomotor responses on probe trials in experiment 1. A–F: force responses for group 1, in which the outward movement was task relevant and the inwardmovement was task irrelevant. A: mean force response to visual perturbations (probe trials) in the normal and task-relevant visual environments across all subjectsin the outward reaching movement. Responses in the first 5 blocks in each condition were not included. The response in the normal environment for both the10% onset (light green) and 30% onset (dark green) perturbations and in the task-relevant visual environment for both the 10% onset (orange) and 30% onset(red) perturbations are shown as a function of time from the onset of the visual perturbation. Colored shaded regions indicate the SE across subjects. Gray barillustrates the involuntary window (180–230 ms). B: mean ! SD force response over the involuntary interval for the outward reaching movements. Statisticallysignificant differences between the conditions were tested with Tukey’s honestly significant difference (HSD) post hoc test (**P % 0.005, ***P % 0.001). N,normal; SD, sensory discrepancy. C: mean force response to 10% onset (light color) and 30% onset (dark color) visual perturbations in the normal (green) andtask-irrelevant (blue) visual environments during inward movements. D: mean ! SD force response over the involuntary interval for the inward movements. E,top: difference in force responses for 10% onset perturbations between the task-dependent environment and the normal environment for both the task-relevant(orange) and task-irrelevant (light blue) movement directions (zero line represents the force response in the normal environment, so that differences can beexamined relative to this response). The difference between these 2 measures was examined every 1 ms with a t-test (Bonferroni corrected for multiplecomparisons) to determine the time point where statistical differences between the 2 task-dependent environments first occurred. Bottom: P values for thiscomparison (black) as well as for each response relative to the normal environment (task relevant: orange; task irrelevant: blue) plotted as a function of the timeafter the onset of the perturbation. Dotted lines show the significance level and the time point at which significance was achieved for each of the comparisons.F, top: difference in force responses for 30% onset perturbations between the task-dependent environment and the normal environment for both the task-relevant(red) and task-irrelevant (dark blue) movement directions. Bottom: P values from the t-test comparisons. G–L: force responses for group 2, in which the outwardmovement was task irrelevant and the inward movement was task relevant.

0

25A B

Posi

tion

Y-ax

is [c

m]

Position X-axis [cm] Position X-axis [cm]

D

-5 0 5

C

-5 0 5-5 0 5-5 0 5

Fig. 5. Kinematics in experiment 2. A: mean cursor trajectories for the 9different trajectories for a single subject in the left-side task-relevant condition.B: mean hand trajectories for the same movements, with the colors correspond-ing to the cursor motion. C: mean cursor trajectories for the 9 differenttrajectories in the right-side task-irrelevant condition. D: mean hand trajecto-ries for the same movements.



on Novem

ber 2, 2014D

ownloaded from

bation size and/or same environment. After a significant maineffect (F3,45 $ 24.1; P % 0.001), differences were examinedwith Tukey’s HSD post hoc test (shown in Fig. 7B). The sameanalysis was performed for the perturbations to the right(negative force responses in Fig. 7A). The ANOVA foundsignificant main effects for both perturbation size (F1,15 $29.989; P % 0.001) and environment (F1,15 $ 12.330; P $

0.003) as well as a significant interaction effect between thesetwo (F1,15 $ 7.554; P $ 0.015). Again, after a significant maineffect across all perturbations (F3,45 $ 16.157; P % 0.001),individual differences were examined with post hoc tests(Tukey’s HSD) as indicated in Fig. 7B.

Identical analysis was performed on the responses on theother day of experiments, where the task-relevant environment

−6 −4 −2 0 2 4 6

0

100

200

300

400

0

100

200

300

400

−6 −4 −2 0 2 4 6

0

100

200

300

400

0

100

200

300

400

−100 −50 0 50 100 −100 −50 0 50 100

A

B

E

F

GC

HD0-5 -2.5 2.5 05 -5 -2.5 2.5 5

−50

0

50

−50

0

50

Cursor position X-axis [cm]

Hand x-acceleration [cm/s2]

Mea

n ha

nd x

-acc

eler

atio

n 15

0-30

0 m

s [c

m/s

2 ]D

iffer

ence

in x

-acc

eler

atio

nfo

r mat

ched

dis

crep

anci

es[c

m/s

2 ]

Cursor discrepancy magnitude [cm] Cursor discrepancy magnitude [cm]

Cursor position X-axis [cm]

Hand x-acceleration [cm/s2]

Tim

e fro

m s

enso

ry

disc

repa

ncy

onse

t [m

s]

Time from sensory discrepancy onset [ms]

Time from sensory discrepancy onset [ms]

Tim

e fro

m s

enso

ry

disc

repa

ncy

onse

t [m

s]

0

5

10

15

20

25

Y P

ositi

on [c

m]

Zero sensory discrepancynull field trials

Zero perturbationchannel field trials

I

0

5

0-1 1 0-2 2

10

15

20

25

X-Position [cm]

Y P

ositi

on [c

m]

X-Force [N]

0-1 1 0-2 2X-Position [cm] X-Force [N]

J

rightwards task relevantleftwards task irrelevant

leftwards task relevantrightwards task irrelevant



no sensorydiscrepancy

sensorydiscrepancy


sensorydiscrepancy

0 200 400

−50

0

50

185 189 197 205

0 200 400

−50

0

50

201199 209 245

5 cm

2.5 cm

1.25 cm

3.75 cm

5 cm

2.5 cm

1.25 cm

3.75 cm



on Novem

ber 2, 2014D

ownloaded from

was on the right-hand side of the reaching movement (Fig. 7,E and F). For perturbations to the left (task-irrelevant side), anANOVA with main effects of perturbation size and environ-mental condition found significant effects for perturbation size(F1,15 $ 39.334; P % 0.001) and environmental condition(F1,15 $ 8.252; P $ 0.012) but no significant interactionbetween these effects (F1,15 $ 1.877; P $ 0.191). The appro-priate individual differences were investigated with a differentANOVA, with significant post hoc results plotted (Fig. 7F)after the significant main effect (F3,45 $ 4.208; P % 0.001). Forperturbations to the right (task-relevant side), an ANOVA withmain effects of perturbation size and environmental conditionfound significant effects of perturbation size (F1,15 $ 39.954;P % 0.001) and environmental condition (F1,15 $ 120.489;P % 0.001) and a significant interaction between these effects(F1,15 $ 13.175; P $ 0.0.002). Again, after a significant maineffect (F3,45 $ 73.999; P % 0.001), the appropriate individualdifferences were investigated with post hoc tests and plotted(Fig. 7F).

The results from the above analysis suggest that the feed-back responses were modulated independently for perturba-tions to the left and right of the movement direction within theinvoluntary time window. However, significant increases werealso found for many of the perturbations into the task-irrelevantside of the environment. To examine whether there was inde-pendent modulation of the feedback responses on each side ofthe movement and determine at what time such independentmodulation occurs, the difference in the force responses be-tween the normal environment and the sensory discrepancyenvironment was calculated every 1 ms within a movement foreach subject (Fig. 7, C, D, G, and H). t-Tests (Bonferronicorrected for multiple comparisons) were used to test whetherthe difference in these responses was significant between theresponses to the left (task relevant) and the right (task irrele-vant) at each time point. Significant differences in the re-sponses were found starting from 245 ms (t1438 $ 3.3894; P $7.19e#4) for the 1.25-cm perturbations (Fig. 7C) and from 225ms (t1438 $ 4.09; P $ 4.57e#5) for the 2.5-cm perturbations(Fig. 7D). Similar times were found for the opposite sensorydiscrepancy environment, where the responses were signifi-cantly larger to the right (task relevant) than to the left (taskirrelevant) (Fig. 7, G and H). Significant differences in theresponses were found starting from 220 ms (t1438 $ 3.74; P $1.89e#4) for the 1.25-cm perturbations and from 205 ms(t1438 $ 3.63; P $ 2.97e#4) for the 2.5-cm perturbations (Fig.7, G and H). These differences (across all subjects) occurred

just prior to the earliest detectable voluntary change in feed-back force found for a single subject (230 ms) (Franklin andWolpert 2008) for such visual perturbations for three of thefour cases. As such, we suggest that these changes inmodulation are occurring within the involuntary responsesrather than through a voluntary correction, although thelow-pass filtering effects of the muscles limit the discrimi-natory power of these responses. To examine this in moredetail, the responses during this involuntary period wereexamined in the muscle activity.

The muscular responses to the visual perturbations of thecursor position were examined in the posterior deltoid andpectoralis major (Fig. 8) and quantified over the involuntarytime period (120–180 ms). Examining the left-side task-rele-vant environment first (Fig. 8, A and B), an ANOVA wasperformed separately for the perturbations to the right and leftfor each muscle. Examining perturbations into the task-rele-vant side of the environment, there was a significant maineffect in the pectoralis major (F3,45 $ 8.15; P % 0.001), withlarger inhibition for the task-relevant compared with normalenvironment in the large perturbation (P % 0.001). In theposterior deltoid, after a significant main effect (F3,45 $14.392; P % 0.001), post hoc analyses demonstrated largerresponses in the task-relevant environment than for the sameperturbation in the normal environment for both perturbationsizes (both P % 0.001). Moreover, in the task-relevant envi-ronment, the 2.5-cm perturbation produced a much largerresponse than the 1.25-cm perturbation (P % 0.001), whereasthere was no significant difference between these two sizes inthe normal environment (P $ 0.452). Perturbations presentedinto the task-irrelevant side of the environment also producedsignificant changes relative to the normal environment. In thepectoralis major, after a significant main effect by ANOVA(F3,45 $ 7.551; P % 0.001), post hoc tests indicated ageneral increase in response for the large perturbation in thetask-irrelevant environment compared with either the largeperturbation in the normal environment (P $ 0.007) or thesmaller perturbation in the task-irrelevant environment (P %0.001). In the posterior deltoid, although there was a sig-nificant main effect (F3,45 $ 4.103; P $ 0.012), post hoctests indicated no significant differences in the relevantcomparisons.

Similar responses can be found in the other environment(right-side task-relevant environment; Fig. 8, C and D). Per-turbations presented into the task-relevant side showed signif-icant differences in the pectoralis major by ANOVA (F3,45 $

Fig. 6. Corrective responses on the sensory discrepancy trials. A: mean (solid lines) and SE across subjects (shaded region) for the cursor position on the sensorydiscrepancy trials for the rightward task-relevant and leftward task-irrelevant environments. Green trace shows the unperturbed mean trajectory for the previousno sensory discrepancy environment. Values are plotted as a function of time from the onset of the sensory discrepancies. B: mean hand acceleration in the lateraldirection for the conditions shown in A. C: mean hand acceleration over the interval 150–300 ms from the onset of the discrepancies. Green line and shadedregion show the mean values for the initial no sensory discrepancy trials. D: difference in the hand x-acceleration between the sensory discrepancies of matchedsize for the conditions in C. The acceleration for the sensory discrepancy to the right (e.g., &5) was contrasted with the acceleration to the left for the samemagnitude discrepancy (e.g., #5) to determine the onset time of any difference in the magnitude of these responses (dashed line). If the magnitudes were of equalsize the difference would remain at zero. Vertical dashed lines indicate the time at which the difference was determined to be significantly different for eachcondition. E: cursor position for the leftward task-relevant and rightward task-irrelevant environments. F: lateral hand acceleration for the condition shown inA. G: mean hand acceleration over the interval 150–300 ms. H: difference in the hand x-acceleration between the sensory discrepancies of matched size for theconditions in G. I, left: comparison of the mean trajectories for the normal reaching trials in which the visual cursor matches the hand location (zero sensorydiscrepancy) in the no sensory discrepancy condition (green) and the rightward task-relevant and leftward task-irrelevant condition (red). Mean, SE, and SD areplotted as a function of the distance to the target. Right: comparison of the mean force exerted on the channel wall in the zero perturbation probe trials in the2 conditions. J, left: mean trajectories for the normal reaching trials (zero sensory discrepancy) in the no sensory discrepancy condition (green) and the rightwardtask-relevant and leftward task-irrelevant condition (red). Right: mean force exerted on the channel wall in the zero perturbation probe trials.



on Novem

ber 2, 2014D

ownloaded from

35.047; P % 0.001). Post hoc analyses demonstrated largerresponses in the task-relevant environment than for the sameperturbation in the normal environment for both perturbationsizes (both P % 0.001). Moreover, in the task-relevant envi-ronment, the 2.5-cm perturbation produced a much largerresponse than the 1.25-cm perturbation (P % 0.001), whereasthere was no significant difference between these two sizes inthe normal environment (P $ 0.869). In the posterior deltoid

no significant changes were found for any of the perturbations(F3,45 $ 1.826; P $ 0.156). However, perturbations presentedinto the task-irrelevant side of the environment also producedsignificant changes relative to the normal environment. In thepectoralis major, after a significant main effect by ANOVA(F3,45 $ 8.105; P % 0.001), post hoc tests indicated a generalincrease in inhibition in the task-irrelevant environment com-pared with the normal environment (both P ! 0.01) but no

**

***

***

***

*****

***-1

0

1

-0.8

0

0.8

0 100 200 300 400

Forc

e [N

]

Forc

e (1

80-2

30m

s) [N

]


A B

1.25 cmProbe

2.5 cmProbe

N SD N SD

taskirrelevant

taskrelevant

taskirrelevant

taskrelevant

-0.8

0

0.8

-1

0

1

0 100 200 300 400

Forc

e [N

]

Forc

e (1

80-2

30m

s) [N

]


E F

1.25 cmProbe

2.5 cmProbe

N SD N SD

***

*********

Forc

e [N

]p-

valu

es

Forc

e [N

]p-

valu

es

1.25 cmProbe

2.5 cmProbe

1.25 cmProbe

2.5 cmProbe


onset [ms]


onset [ms]


onset [ms]


onset [ms]

−1

0

1

−1

0

1

−1

0

1

−1

0

1

C D

G H

10-20

10-10

100

10-20

10-10

100

10-2010-20

10-10

100

10-10

100

Forc

e [N

]p-

valu

es

Forc

e [N

]p-

valu

es

004001004001

004001004001

p=1.7x10-4

p=1.7x10-4

502022

522542

Fig. 7. Visuomotor responses on probe trials in experiment 2 in both the normal and sensory discrepancy environments. A: mean force response to the small (1.25cm) and large (2.5 cm) visual perturbations in the normal and right-side task-relevant environments across all subjects. The first 5 blocks in each condition werenot included. The colors correspond to the environment (normal: green; task relevant: red; and task irrelevant: blue) for both the 1.25-cm (light colors) and 2.5-cm(dark colors) probe trials as a function of time after perturbation onset. B: mean ! SD force response over the involuntary interval (180–230 ms). Statisticallysignificant differences between the conditions were tested with Tukey’s HSD post hoc test (**P % 0.005; ***P % 0.001). C, top: difference in force responsesfor 1.25-cm perturbations between the sensory discrepancy environment and the normal environment for both the task-relevant (orange) and task-irrelevant (lightblue) perturbation directions. The difference between these 2 measures was examined every 1 ms with a t-test (Bonferroni corrected for multiple comparisons)to determine the time point where differences between the 2 perturbation directions first occurred. Bottom: P values for these comparisons plotted as a functionof the time after the onset of the perturbation. Dotted line shows the level at which significance was considered. D: difference in force responses for 2.5-cmperturbations between the sensory discrepancy environment and the normal environment for both the task-relevant (red) and task-irrelevant (dark blue)perturbation directions. E: mean force responses for the left-side task-relevant environment experiment. F: mean ! SD force responses across subjects over theinvoluntary interval. G: change in force responses between the task-relevant and task-irrelevant perturbations relative to the original normal environmentfor the 1.25-cm probe trials. H: change in force responses between the task-relevant and task-irrelevant perturbations relative to the original normalenvironment for the 2.5-cm probe trials.



on Novem

ber 2, 2014D

ownloaded from

changes between the small and large perturbation sizes ineither environment (both P ' 0.35). In the posterior deltoid,after a significant main effect (F3,45 $ 6.53; P $ 0.001), posthoc tests indicated an increase in the response to the twoperturbation sizes in the task-irrelevant environment (P %0.001), with no other significant differences in the appropriate

comparisons. Overall, although significant effects are seen inthese early involuntary intervals for both the task-relevant andtask-irrelevant environments, the difference between these two,with increased excitation and consistent and appropriatechanges in feedback gain in the task-relevant environment, canbe clearly seen in the responses.

1.25 cm 2.5 cm

taskrelevant

1.25 cm 2.5 cmnormal

1.25 cm 2.5 cm

taskirrelevant

0 100 2000

2

4

6

0 100 2000

2

4

6

0 100 2000

2

4

6

0 100 2000

2

4

6



DC no sensorydiscrepancy


Cha

nge

in P

ecto

ralis

Maj

orAc

tivity

(120

-180

ms)

Cha

nge

in P

oste

rior D

elto

idAc

tivity

(120

-180

ms)

−1

0

1

2

3

0

2

4

6

1.25 cmProbe

2.5 cmProbe

N SD N SD

***

***

***

***

******



BA

Pect

oral

is M

ajor

[a.u

]Po

ster

ior D

elto

id[a

.u]


Cha

nge

in P

ecto

ralis

Maj

orAc

tivity

(120

-180

ms)

Cha

nge

in P

oste

rior D

elto

idAc

tivity

(120

-180

ms)


0 100 2000

2

4

6

0 100 2000

2

4

6

0 100 2000

2

4

6

0 100 2000

2

4

6

1.25 cmProbe

2.5 cmProbe

N SD N SD

−1

0

1

2

3

0

2

4

6

*****

***

******

***

Pect

oral

is M

ajor

[a.u

]Po

ster

ior D

elto

id[a

.u]

Fig. 8. Muscular responses to visual perturbations(probe trials) during experiment 2. A: mean (solidline) ! SE (shaded region) muscle activity [arbitraryunits (a.u.)] in the pectoralis major (top) and posteriordeltoid (bottom) for perturbations in the normal (left)and leftward task-relevant sensory discrepancy (right)environments. Responses to the perturbations into thetask-relevant part of the environment are shown inorange (1.25 cm) and red (2.5 cm), into the task-irrelevant part of the environment in light (1.25 cm) anddark (2.5 cm) blue, and into the normal environment inlight (1.25 cm) and dark (2.5 cm) green. The response tothe zero visual perturbation condition is shown in black.Shaded gray bar indicates the involuntary response in-terval (120–180 ms). B: mean ! SD muscular responseover the involuntary interval (180–230 ms) for theleftward task-relevant sensory discrepancy environ-ment. Statistically significant differences between theconditions were tested with Tukey’s HSD post hoc test(**P % 0.005; ***P % 0.001). C: mean (solid line) !SE (shaded region) muscle activity in the pectoralismajor (top) and posterior deltoid (bottom) for perturba-tions in the normal (left) and rightward task-relevantsensory discrepancy (right) environments. D: mean !SD muscular response over the involuntary interval forthe rightward task-relevant sensory discrepancy envi-ronment. ***P % 0.001.



on Novem

ber 2, 2014D

ownloaded from

DISCUSSION

The results of this study provide clear evidence that the braincan learn to modulate the gain of the visuomotor response(motor output to perturbations of the visual location of the handduring movement) for a single limb posture depending on thedirection of either the movement or the perturbation. Specifi-cally, we have exploited the finding that this visuomotorresponse magnitude is affected by the statistical properties ofthe visual environment (Franklin and Wolpert 2008) to deter-mine whether the visuomotor responses could be indepen-dently modulated either to the direction of hand motion or tothe direction of the perturbation. The first experiment demon-strated that subjects could appropriately learn to produce largerfeedback responses in outward reaching movements whilealternating with smaller feedback responses in the inwardmovements (or vice versa) when these two movement direc-tions were paired with task-relevant and task-irrelevant pertur-bations, respectively. The second experiment demonstratedthat the feedback responses could be independently tuned tothe left and right of the movement, with gain increasing ononly one side in which task-relevant perturbations occurred.We suggest that these results indicate that subjects are able tolearn to fractionate their feedback responses as a function ofthe state of the limb and the state of the perturbation.

The first experiment showed differential feedback modula-tion for alternating reaches in two directions. Interestingly, theearly feedback responses in the task-irrelevant condition werenot significantly different from those in the normal environ-ment, in contrast to our previous study (Franklin and Wolpert2008). One explanation is that the normal environment may beconsidered task irrelevant, as the random probe trials containvisual perturbations that themselves are a type of sensorydiscrepancy. However, this is true for both the present studyand the 2008 study, so it is unlikely to contribute to thedifference between them. It has previously been shown thatlearning to inhibit the visuomotor feedback gain is muchslower than increasing the gain (Franklin and Wolpert 2008)and produces a smaller difference. Thus any competition (orinterference) between excitation and inhabitation of the feed-back pathways in the alternating movements would produce alarger influence on the responses in the task-irrelevant condi-tion. This experiment also produced responses to the task-relevant condition that appeared to lead the normal responsesby a fixed delay rather than a simple gain increase under some(but not all) conditions (Fig. 4). Although there are variationsin the response onset time (Fig. 4), these were not consistentacross conditions, suggesting that it is unlikely to result fromactual differences in the latencies. Moreover, any significantearly difference was maintained throughout the response time,indicating that this difference was maintained. Any differencesin these responses (either in timing or magnitude) are unlikelyto be due to an effect of either attention or a startlelikeresponse. This is because attention has been shown not tomodulate the responses to cursor perturbations (Reichenbach etal. 2014). Similarly, startlelike responses, if induced, wouldfirst contribute to changes in the force in 105 ms (Valls-Solé etal. 1995) rather than 140 ms and, more critically, should beinduced across all conditions (as the perturbation trials areidentical) rather than being expressed only in a task-dependentmanner based on the surrounding nonprobe trials.

The visuomotor feedback responses that in this study rangefrom 0.5 to 1 N have been shown to vary in magnitude up to1.5 N (Dimitriou et al. 2013; Franklin et al. 2012; Franklin andWolpert 2008; Reichenbach et al. 2013, 2014) depending onthe particular environment in which subjects are moving andthe task they are performing. Although these magnitudes mayappear small, it is worth considering that they contribute achange in EMG activity and end-point force similar to thosearising from stretch reflexes (Franklin et al. 2008). Moreover,even the addition of visuomotor responses, stretch reflex re-sponses, and limb stiffness will produce between 3 and 5 N offorce in response to perturbations of 8 mm during whole-armreaching movements (Burdet et al. 2000). This can also beestimated from the measured end-point stiffness of the armduring null-field reaching movements, which ranges in thelateral direction from 200 to 400 N/m (Franklin et al. 2003,2004, 2007). A perturbation of 1 cm would therefore beexpected to result in a change in force of 2–4 N total from allcontributions including muscle stiffness and feedback re-sponses. Thus even this relatively small change in end-pointforce produces a significant contribution of the total responseduring a reaching movement, and its effect on the movement ofthe limb can be seen both in the acceleration changes on thesensory discrepancy trials (Fig. 6) and in other studies (Brennerand Smeets 2003; Wijdenes et al. 2011, 2013).

There are extensive studies demonstrating the modulation offeedback responses to the properties of the task that subjectsperform (Franklin and Wolpert 2011; Pruszynski and Scott2012). For example, studies have demonstrated the modulationof both stretch reflex and visuomotor responses to changes inthe environmental dynamics (Akazawa et al. 1983; Cluff andScott 2013; Doemges and Rack 1992a, 1992b; Franklin et al.2012; Kimura and Gomi 2009), the visual statistics (Franklinand Wolpert 2008), the shape of the target (Knill et al. 2011;Nashed et al. 2012), and when the two limbs manipulate asingle object (Diedrichsen 2007; Dimitriou et al. 2012; Omraniet al. 2013). Similarly, studies have shown that the long-latency stretch responses are modulated appropriately to thelimb dynamics (Kurtzer et al. 2008; Lacquaniti and Soechting1986; Pruszynski et al. 2011a; Soechting and Lacquaniti 1988),producing required compensatory responses even when themuscle is not perturbed. However, most of these studieshave examined the feedback responses for a particular task,where during the time period that the subject performs thetask the feedback gains are either globally increased ordecreased appropriately.

Very few studies have examined the changes in feedbackresponses within a single task, i.e., simultaneously increasingand decreasing the feedback gains appropriately within a singlemovement or task. Similar to experiments in locomotion wherethe gains vary with the phases of the step cycle (Capaday andStein 1986; Sinkjaer et al. 1996), it has been shown that thevisuomotor gains vary as a function of the distance to the target(Dimitriou et al. 2013; Liu and Todorov 2007; Wijdenes et al.2011). All of these responses vary according to the taskrequirements (stage of locomotion or distance to target), mean-ing that they also vary with the limb posture. It is possible,therefore, to consider most of these state-dependent results astask-dependent expression of the spinal circuitry (Pearson2000; Schieppati 1987). In contrast, here we have shown notonly that these responses are prestored functional responses



on Novem

ber 2, 2014D

ownloaded from

varying only with the limb posture but that they can be learnedappropriately for the particular environment, tuned indepen-dently (upregulated and downregulated) for identical limbpostures with different velocity or perturbation direction states.This demonstrates that visuomotor feedback gains can also belearned independently to either side of the movement, similarto what has been previously demonstrated for stretch feedbackresponses when learning to compensate for novel dynamics(Kimura and Gomi 2009) or to avoid objects during reaching(Nashed et al. 2014). This work also extends the findings ofKnill et al. (2011), showing that these feedback gains can bemodulated differentially to either side of the movement and notonly between perturbations in direction and extent.

Previous studies of visuomotor feedback responses haveshown clear changes in the feedback magnitudes from theearliest detectable responses (Franklin et al. 2012; Franklin andWolpert 2008). For example, when the sensory discrepancywas either task relevant or task irrelevant, the differences wereapparent from the initial response (140 ms onward), leading toclear differences over the involuntary window used in thisstudy (Franklin and Wolpert 2008). However, in this study,although the differences between the task-relevant and task-irrelevant conditions were clear at later intervals, the earliestresponses were not statistically different between these twoconditions. It is only closer to 180 ms (experiment 1) or 200 ms(experiment 2) that these differences become statically signif-icant. This suggests that fractionating the response based onmovement or perturbation direction may require extra process-ing time. Such a possibility is similar in some sense to the rapidmotor responses to muscle stretch: different functional re-sponses at different delay times (Crago et al. 1976; Pruszynskiet al. 2011b). For stretch responses, although the initial short-latency response is not modifiable [except on a long timescale(Wolpaw et al. 1983)] and only exhibits gain scaling (Marsdenet al. 1976; Pruszynski et al. 2009), the longer-latency re-sponses can be modulated according to the task demands(Dimitriou et al. 2012; Kurtzer et al. 2008; Nashed et al. 2012,2014; Shemmell et al. 2009) and may involve cortical circuitssimilar to those involved in voluntary motor control (Pruszyn-ski et al. 2011a; Zuur et al. 2009, 2010) as well as subcorticalcomponents (Grey et al. 2001; Lewis et al. 2004).

However, unlike the stretch reflex, the pathways involved inproducing the rapid visuomotor responses are unknown. Someevidence suggests the involvement of the superior colliculus,which is strongly involved in visually guided motion (Courjonet al. 2004; Gahtan et al. 2005; Song et al. 2011; Sprague andMeikle 1965), whereas other evidence has suggested the pre-motor cortex, specifically PMv (Jackson and Husain 1996;Kurata 1994) or parietal cortex (Diedrichsen et al. 2005). Onepossibility is that the later differentiation of the feedback gainsinvolves a neural circuit different from the earliest response,either allowing for further processing of the information orintegrating further sensory feedback and task goals into theresponse. However, it is important to note that the responsesduring this interval (up to 230 ms) still cannot be eliminateddespite explicit task commands (Day and Lyon 2000; Franklinand Wolpert 2008).

The subjects in experiment 2 not only modified their feed-back responses but also shifted their mean trajectory toward thetask-irrelevant side of the sensory discrepancy environment.This change in the movement execution acted to both partially

compensate for the possible maintained task-relevant perturba-tions (therefore requiring less corrective response) and partiallycompensate (in advance of the perturbation) for the elicitedfeedback response in the task-irrelevant perturbations. Theonly condition in which this change in trajectory was notappropriate was the zero sensory discrepancy trial. Thus thesubjects found a shift that helped to compensate but stillallowed this trial to end in the target location. We believe thisshift in the trajectory is another example of the sensorimotorcontrol system reoptimizing the trajectory to ensure task com-pletion (Izawa et al. 2008; Nagengast et al. 2009).

One of the key results of this work is the demonstration thatthese rapid feedback responses can be adapted simultaneouslyto multiple environmental conditions during a single move-ment. This demonstrates several important features of thefeedback gains. The first feature is that these feedback gainsare continually learned and adapted to the tasks being per-formed. Indeed, several studies have shown the gradual learn-ing of feedback gains as subjects learn environmental dynam-ics (Cluff and Scott 2013; Franklin et al. 2012). The secondfeature is that these responses can be fractionated appropriatelyrather than simply upregulated or downregulated. Although ona single movement only one feedback gain was actually ex-pressed (because of our applied perturbation), the feedbackgains for the left and right visuomotor responses had to beappropriately set, as the direction of the perturbation was notknown in advance (experiment 2). The sensorimotor controlsystem can, therefore, prepare an appropriate pattern of feed-back gains for the environment and tune these feedback re-sponses to the environment as part of the adaptation process forthe performance of skilled movement.

ACKNOWLEDGMENTS

We thank J. N. Ingram and I. S. Howard for their work in setting up thevBOT robotic interface used in this study.

GRANTS

This work was financially supported by the Wellcome Trust (WT091547MAand WT097803MA), the Biotechnology and Biological Sciences Research Coun-cil (BBSRC; BB/ J009458/1) and the Royal Society Noreen Murray Professor-ship in Neurobiology (to D. M. Wolpert).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: D.W.F. and D.M.W. conception and design of re-search; D.W.F. and S.F. performed experiments; D.W.F. analyzed data;D.W.F. interpreted results of experiments; D.W.F. prepared figures; D.W.F.drafted manuscript; D.W.F., S.F., and D.M.W. edited and revised manuscript;D.W.F., S.F., and D.M.W. approved final version of manuscript.

REFERENCES

Ahmadi-Pajouh MA, Towhidkhah F, Shadmehr R. Preparing to reach:selecting an adaptive long-latency feedback controller. J Neurosci 32:9537–9545, 2012.

Akazawa K, Milner TE, Stein RB. Modulation of reflex EMG and stiffnessin response to stretch of human finger muscle. J Neurophysiol 49: 16–27,1983.

Brenner E, Smeets JB. Fast corrections of movements with a computermouse. Spat Vis 16: 365–376, 2003.



on Novem

ber 2, 2014D

ownloaded from

Burdet E, Osu R, Franklin D, Yoshioka T, Milner T, Kawato M. A methodfor measuring endpoint stiffness during multi-joint arm movements. JBiomech 33: 1705–1709, 2000.

Capaday C, Stein RB. Amplitude modulation of the soleus H-reflex in thehuman during walking and standing. J Neurosci 6: 1308–1313, 1986.

Cluff T, Scott SH. Rapid feedback responses correlate with reach adaptationand properties of novel upper limb loads. J Neurosci 33: 15903–15914,2013.

Corneil BD, Munoz DP, Chapman BB, Admans T, Cushing SL. Neuro-muscular consequences of reflexive covert orienting. Nat Neurosci 11:13–15, 2007.

Corneil BD, Olivier E, Munoz DP. Visual responses on neck muscles revealselective gating that prevents express saccades. Neuron 42: 831–841, 2004.

Courjon JH, Olivier E, Pelisson D. Direct evidence for the contribution of thesuperior colliculus in the control of visually guided reaching movements inthe cat. J Physiol 556: 675–681, 2004.

Crago PE, Houk JC, Hasan Z. Regulatory actions of human stretch reflex. JNeurophysiol 39: 925–935, 1976.

Day B, Lyon I. Voluntary modification of automatic arm movements evokedby motion of a visual target. Exp Brain Res 130: 159–168, 2000.

Diedrichsen J. Optimal task-dependent changes of bimanual feedback controland adaptation. Curr Biol 17: 1675–1679, 2007.

Diedrichsen J, Hashambhoy Y, Rane T, Shadmehr R. Neural correlates ofreach errors. J Neurosci 25: 9919–9931, 2005.

Dimitriou M, Franklin DW, Wolpert DM. Task-dependent coordination ofrapid bimanual motor responses. J Neurophysiol 107: 890–901, 2012.

Dimitriou M, Wolpert DM, Franklin DW. The temporal evolution offeedback gains rapidly update to task demands. J Neurosci 33: 10898–10909, 2013.

Doemges F, Rack PM. Changes in the stretch reflex of the human first dorsalinterosseous muscle during different tasks. J Physiol 447: 563–573, 1992a.

Doemges F, Rack PM. Task-dependent changes in the response of humanwrist joints to mechanical disturbance. J Physiol 447: 575–585, 1992b.

Franklin DW, Burdet E, Osu R, Kawato M, Milner TE. Functionalsignificance of stiffness in adaptation of multijoint arm movements to stableand unstable dynamics. Exp Brain Res 151: 145–157, 2003.

Franklin DW, Liaw G, Milner TE, Osu R, Burdet E, Kawato M. Endpointstiffness of the arm is directionally tuned to instability in the environment.J Neurosci 27: 7705–7716, 2007.

Franklin DW, So U, Kawato M, Milner TE. Impedance control balancesstability with metabolically costly muscle activation. J Neurophysiol 92:3097–3105, 2004.

Franklin DW, So U, Osu R, Kawato M. Conflicting visual and propriocep-tive reflex responses during reaching movements. In: Neural InformationProcessing, Lecture Notes in Computer Science, edited by Ishikawa M,Doya K, Miyamoto H, Yamakawa T. Berlin: Springer, 2008, p. 1002–1011.

Franklin DW, Wolpert DM. Specificity of reflex adaptation for task-relevantvariability. J Neurosci 28: 14165–14175, 2008.

Franklin DW, Wolpert DM. Computational mechanisms of sensorimotorcontrol. Neuron 72: 425–442, 2011.

Franklin S, Wolpert DM, Franklin DW. Visuomotor feedback gains up-regulate during the learning of novel dynamics. J Neurophysiol 108: 467–478, 2012.

Gahtan E, Tanger P, Baier H. Visual prey capture in larval zebrafish iscontrolled by identified reticulospinal neurons downstream of the tectum. JNeurosci 25: 9294–9303, 2005.

Gomi H, Abekawa N, Nishida S. Spatiotemporal tuning of rapid interactionsbetween visual-motion analysis and reaching movement. J Neurosci 26:5301–5308, 2006.

Goodale MA, Pelisson D, Prablanc C. Large adjustments in visually guidedreaching do not depend on vision of the hand or perception of targetdisplacement. Nature 320: 748–750, 1986.

Grey MJ, Ladouceur M, Andersen JB, Nielsen JB, Sinkjaer T. Group IImuscle afferents probably contribute to the medium latency soleus stretchreflex during walking in humans. J Physiol 534: 925–933, 2001.

Howard IS, Ingram JN, Wolpert DM. A modular planar robotic manipu-landum with end-point torque control. J Neurosci Methods 181: 199–211,2009.

Izawa J, Rane T, Donchin O, Shadmehr R. Motor adaptation as a process ofreoptimization. J Neurosci 28: 2883–2891, 2008.

Jackson SR, Husain M. Visuomotor functions of the lateral pre-motor cortex.Curr Opin Neurobiol 6: 788–795, 1996.

Kimura T, Gomi H. Temporal development of anticipatory reflex modulationto dynamical interactions during arm movement. J Neurophysiol 102:2220–2231, 2009.

Knill DC, Bondada A, Chhabra M. Flexible, task-dependent use of sensoryfeedback to control hand movements. J Neurosci 31: 1219–1237, 2011.

Kurata K. Information processing for motor control in primate premotorcortex. Behav Brain Res 61: 135–142, 1994.

Kurtzer IL, Pruszynski JA, Scott SH. Long-latency reflexes of the humanarm reflect an internal model of limb dynamics. Curr Biol 18: 449–453,2008.

Lacquaniti F, Soechting JF. EMG responses to load perturbations of theupper limb: effect of dynamic coupling between shoulder and elbow motion.Exp Brain Res 61: 482–496, 1986.

Lewis GN, Polych MA, Byblow WD. Proposed cortical and sub-corticalcontributions to the long-latency stretch reflex in the forearm. Exp Brain Res156: 72–79, 2004.

Liu D, Todorov E. Evidence for the flexible sensorimotor strategies predictedby optimal feedback control. J Neurosci 27: 9354–9368, 2007.

Marsden CD, Merton PA, Morton HB. Servo action in the human thumb. JPhysiol 257: 1–44, 1976.

Milner TE, Franklin DW. Impedance control and internal model use duringthe initial stage of adaptation to novel dynamics in humans. J Physiol 567:651–664, 2005.

Nagengast AJ, Braun DA, Wolpert DM. Optimal control predicts humanperformance on objects with internal degrees of freedom. PLoS Comput Biol5: e1000419, 2009.

Nashed JY, Crevecoeur F, Scott SH. Influence of the behavioral goal andenvironmental obstacles on rapid feedback responses. J Neurophysiol 108:999–1009, 2012.

Nashed JY, Crevecoeur F, Scott SH. Rapid online selection between multiplemotor plans. J Neurosci 34: 1769–1780, 2014.

Oldfield RC. The assessment and analysis of handedness: the Edinburghinventory. Neuropsychologia 9: 97–113, 1971.

Omrani M, Diedrichsen J, Scott SH. Rapid feedback corrections during abimanual postural task. J Neurophysiol 109: 147–161, 2013.

Pearson KG. Neural adaptation in the generation of rhythmic behavior. AnnuRev Physiol 62: 723–753, 2000.

Prablanc C, Martin O. Automatic control during hand reaching at undetectedtwo-dimensional target displacements. J Neurophysiol 67: 455–469, 1992.

Pruszynski JA, Kurtzer I, Lillicrap TP, Scott SH. Temporal evolution of“automatic gain-scaling.” J Neurophysiol 102: 992–1003, 2009.

Pruszynski JA, Kurtzer I, Nashed JY, Omrani M, Brouwer B, Scott SH.Primary motor cortex underlies multi-joint integration for fast feedbackcontrol. Nature 478: 387–390, 2011a.

Pruszynski JA, Kurtzer I, Scott SH. The long-latency reflex is composed ofat least two functionally independent processes. J Neurophysiol 106: 449–459, 2011b.

Pruszynski JA, Scott SH. Optimal feedback control and the long-latencystretch response. Exp Brain Res 218: 341–359, 2012.

Reichenbach A, Costello A, Zatka-Haas P, Diedrichsen J. Mechanisms ofresponsibility assignment during redundant reaching movements. J Neuro-physiol 109: 2021–2028, 2013.

Reichenbach A, Franklin DW, Zatka-Haas P, Diedrichsen J. A dedicatedbinding mechanism for the visual control of movement. Curr Biol 24:780–785, 2014.

Saijo N, Murakami I, Nishida S, Gomi H. Large-field visual motion directlyinduces an involuntary rapid manual following response. J Neurosci 25:4941–4951, 2005.

Sarlegna F, Blouin J, Bresciani JP, Bourdin C, Vercher JL, Gauthier GM.Target and hand position information in the online control of goal-directedarm movements. Exp Brain Res 151: 524–535, 2003.

Saunders JA, Knill DC. Humans use continuous visual feedback from thehand to control fast reaching movements. Exp Brain Res 152: 341–352,2003.

Scheidt RA, Reinkensmeyer DJ, Conditt MA, Rymer WZ, Mussa-IvaldiFA. Persistence of motor adaptation during constrained, multi-joint, armmovements. J Neurophysiol 84: 853–862, 2000.

Schieppati M. The Hoffmann reflex: a means of assessing spinal reflexexcitability and its descending control in man. Prog Neurobiol 28: 345–376,1987.

Scott SH. Optimal feedback control and the neural basis of volitional motorcontrol. Nat Rev Neurosci 5: 532–546, 2004.



on Novem

ber 2, 2014D

ownloaded from

Shemmell J, An JH, Perreault EJ. The differential role of motor cortex instretch reflex modulation induced by changes in environmental mechanicsand verbal instruction. J Neurosci 29: 13255–13263, 2009.

Sinkjaer T, Andersen JB, Larsen B. Soleus stretch reflex modulation duringgait in humans. J Neurophysiol 76: 1112–1120, 1996.

Soechting JF, Lacquaniti F. Quantitative evaluation of the electromyographicresponses to multidirectional load perturbations of the human arm. J Neu-rophysiol 59: 1296–1313, 1988.

Song JH, Rafal RD, McPeek RM. Deficits in reach target selection duringinactivation of the midbrain superior colliculus. Proc Natl Acad Sci USA108: E1433–E1440, 2011.

Sprague JM, Meikle TH. The role of the superior colliculus in visuallyguided behavior. Exp Neurol 11: 115–146, 1965.

Todorov E. Optimality principles in sensorimotor control. Nat Neurosci 7:907–915, 2004.

Todorov E, Jordan MI. Optimal feedback control as a theory of motorcoordination. Nat Neurosci 5: 1226–1235, 2002.

Valls-Solé J, Solé A, Valldeoriola F, Muñoz E, Gonzalez LE, Tolosa ES.Reaction time and acoustic startle in normal human subjects. Neurosci Lett195: 97–100, 1995.

Wijdenes LO, Brenner E, Smeets JB. Fast and fine-tuned corrections whenthe target of a hand movement is displaced. Exp Brain Res 214: 453–462,2011.

Wijdenes LO, Brenner E, Smeets JB. Comparing online adjustments todistance and direction in fast pointing movements. J Mot Behav 45:395–404, 2013.

Wolpaw JR, Braitman DJ, Seegal RF. Adaptive plasticity in primate spinalstretch reflex: initial development. J Neurophysiol 50: 1296–1311, 1983.

Zuur AT, Christensen MS, Sinkjaer T, Grey MJ, Nielsen JB. Tibialisanterior stretch reflex in early stance is suppressed by repetitive transcranialmagnetic stimulation. J Physiol 587: 1669–1676, 2009.

Zuur AT, Lundbye-Jensen J, Leukel C, Taube W, Grey MJ, Gollhofer A,Nielsen JB, Gruber M. Contribution of afferent feedback and descendingdrive to human hopping. J Physiol 588: 799–807, 2010.



on Novem

ber 2, 2014D

ownloaded from

Fractionation of the visuomotor feedback response to ......2013/05/23 · cally, task-relevant...

Documents

Transcript of Fractionation of the visuomotor feedback response to ......2013/05/23 · cally, task-relevant...